A Wiener filter approach to coherence estimation

This paper presents a new method of estimating the magnitude-squared-coherence (MSC) between two random processes and . An unrealizable Wiener filter transfer function is first computed in the channel, then another one in the y channel. The product of these transfer functions then gives the MSC. The unrealizable Wiener filter is approximated by a finite impulse filter whose coefficients can be computed from any standard parameter estimation algorithm. Theoretical upper bounds on the variances of the MSC estimates, together with experimental results, are given.